Preventing Directory Traversal in PHP but allowing paths

I have a base path /whatever/foo/

and $_GET['path'] should be relative to it.

However how do I accomplish this (reading the directory), without allowing directory traversal?

eg.

/\.\.|\.\./

Will not filter properly.

Solution 1:

Well, one option would be to compare the real paths:

$basepath = '/foo/bar/baz/';
$realBase = realpath($basepath);

$userpath = $basepath . $_GET['path'];
$realUserPath = realpath($userpath);

if ($realUserPath === false || strpos($realUserPath, $realBase) !== 0) {
    //Directory Traversal!
} else {
    //Good path!
}

Basically, realpath() will resolve the provided path to an actual hard physical path (resolving symlinks, .., ., /, //, etc)... So if the real user path does not start with the real base path, it is trying to do a traversal. Note that the output of realpath will not have any "virtual directories" such as . or .....

Solution 2:

ircmaxell's answer wasn't fully correct. I've seen that solution in several snippets but it has a bug which is related to the output of realpath(). The realpath() function removes the trailing directory separator, so imagine two contiguous directories such as:

/foo/bar/baz/

/foo/bar/baz_baz/

As realpath() would remove the last directory separator, your method would return "good path" if $_GET['path'] was equal to "../baz_baz" as it would be something like

strpos("/foo/bar/baz_baz", "/foo/bar/baz")

Maybe:

$basepath = '/foo/bar/baz/';
$realBase = realpath($basepath);

$userpath = $basepath . $_GET['path'];
$realUserPath = realpath($userpath);

if ($realUserPath === false || strcmp($realUserPath, $realBase) !== 0 || strpos($realUserPath, $realBase . DIRECTORY_SEPARATOR) !== 0) {
    //Directory Traversal!
} else {
    //Good path!
}

Solution 3:

It is not sufficient to check for patterns like ../ or the likes. Take "../" for instance which URI encodes to "%2e%2e%2f". If your pattern check happens before a decode, you would miss this traversal attempt. There are some other tricks hackers can do to circumvent a pattern checker especially when using encoded strings.

I've had the most success stopping these by canonicalizing any path string to its absolute path using something like realpath() as ircmaxwell suggests. Only then do I begin checking for traversal attacks by matching them against a base path I've predefined.

Solution 4:

You may be tempted to try and use regex to remove all ../s but there are some nice functions built into PHP that will do a much better job:

$page = basename(realpath($_GET));

basename - strips out all directory information from the path e.g. ../pages/about.php would become about.php

realpath - returns a full path to the file e.g. about.php would become /home/www/pages/about.php, but only if the file exists.

Combined they return just the files name but only if the file exists.